JFR recording footprint

[grzesuav]

hi, in the section https://cryostat.io/guides/#archive-a-recording I see the following statement

In contrast to Active Recordings, which reside within the container of the target application, Archived Recordings reside within persistent storage attached to a Cryostat instance

what is the actual footprint of active recording? I mean, in case when I have continous profiling enabled, does cryostat is pulling the data within some period etc ?

[andrewazores]

Cryostat will only pull the data periodically if you have created an Automated Rule to do so. Or, if you are using the Agent, then there is the Harvester feature where it will push the data to Cryostat periodically. If you simply start a recording manually then that data will never be pulled out of the target JVM until you click the "archive" button.

The actual footprint of the recording within the target JVM is whatever (usually quite small) CPU overhead there is for the JVM to capture the events, and whatever memory is taken up by the JFR events in the JVM's memory buffers. The JVM will flush these buffers to disk so long as the "to disk" recording option is enabled (which is on by default), however if your container is configured so that the default write location is in tmpfs or some other volatile medium then this may actually end up still taking memory allocation from the container.

[grzesuav]

@andrewazores thanks, and regarding disk footprint - in case when I started continuous recording on app -

so the JFR sample in the container will (as of disk size) fill up till ephemeral disk space / mem if I got it right ? I was thinking it can be somehow configured, like take up to 50m, then remove oldest data

[andrewazores]

You may find these resources helpful:

https://docs.oracle.com/javacomponents/jmc-5-3/jfr-runtime-guide/about.htm#CHDHAGAF

https://docs.oracle.com/javacomponents/jmc-5-4/jfr-runtime-guide/run.htm#JFRUH164

When you enable JFR, the JVM keeps some in-memory buffers to hold the JFR events. If you enable the todisk option then the JVM will flush the data to disk storage periodically, as the in-memory buffers fill up. You can also set the maximum age and/or size of data that should be kept in this on-disk "JFR repository", so that sounds like it would satisfy your "take up to 50m" idea.

Under the Advanced Options in the Cryostat UI you can configure these things. The todisk is on by default, so you can set your 50MB limit there as well if that suits you.